See you in 2026. 🤝 @GoogleAI 🤝 @GoogleResearch 🤝 @GoogleQuantumAI

imo this is a HUGE deal Google DeepMind and the UK govt have just signed a partnership w/ 3 pillars: 1. transforming public services 2. accelerating scientific discovery 3. advancing AI security & resilience Concrete bits that matter: - a Gemini model trained on the UK

Google is the only company that owns the entire AI stack. 1. Chips: TPUs 2. Infrastructure: Google Cloud 3. Foundation Models: Deepmind 4. Applications: Gemini Every layer matters. The more layers you own, the bigger your advantage.

1. In a podcast, LeCun stated that the non-existence of "general intelligence" is utter nonsense. He argued that human intelligence is highly specialized, only "general" in conceivable problems, and has a vast number of unimaginable tasks. 2. DeepMind's CEO, Hassabis, directly

Mid 2022 - everyone laughed at the Google autist for developing what we now know as LLM psychosis. End of 2025 - a former DeepMind Engineering director has been convinced of something very improbable… 🫣 2026 is going to be interesting.

Our team at Google DeepMind in Toronto is hiring student researchers for 2026 to work on cutting-edge vision research, focusing on both generative models and scene understanding. Come work with us! Apply at: https://t.co/YBB3lnTqwl

This guy has 15,000 citations and was a lead of engineering at Deepmind. You really can just do things 💀

an ex-director at deepmind, exhibit B: just a guy who waves credentials when he supposedly doesn't care about them, claims to be an artist, counts likes and impressions against an early-career researcher, and shushes people "as an adult"

Came third place at the DeepMind Gemini Hackathon in London :)

Google DeepMind is releasing Gemma Scope 2: SAEs and transcoders on every layer of every Gemma 3 model, 270M-27B, base & chat. We hope this enables deep dives into complex model behavior, for more ambitious open source safety & interpretability work!

🚨 @cHHillee followed @davidmbudden Former Google DeepMind Director of Engineering @davidmbudden founded PingYou, an app building real connections for solo travelers and introverts.

Google DeepMind releases FunctionGemma, a 240M model specialized in 🔧 tool calling, built for fine-tuning TRL has day-0 support. To celebrate, we’re sharing new resources: - Colab guide to fine-tune it for 🌐 browser control with BrowserGym OpenEnv - Standalone training script

grrr I really want to apply to this summer DeepMind internship but despite satisfying a bunch of the eligibility criteria, I can't apply to it because it's NOT open to Cambridge students... SO WHY IS IT LISTED ON THE CAMBRIDGE SITE THEN GRAHHH

🧪 What the FACTS Benchmark Suite Is ? Google DeepMind, in collaboration with Kaggle, introduced the FACTS Benchmark Suite to systematically evaluate factual accuracy in LLMs. It tests models across parametric knowledge, web search, multimodal understanding, and grounding.

New Deepmind paper on the science of agentic systems 👀 As much as there's talk about post training right now, simply writing the right agentic harness gets you a lot of early wins as a product. This paper gives a scalable recipe for what type of agent harnesses to use for what

Fine tune Google's FunctionGemma for Mobile, with agents, on colab, locally, or Hugging Face. Google Deepmind Have just release FunctionGemma and anyone can finetune it with TRL. This is the model: - uses the Gemma 3 270M architecture + adapted chat template - specifically for

笑死，GPT-5.2 刚发布就被突脸 GPT-5.2 在发布上展示了一个电脑主板上面的元件和接口识别标记的case，来展示GPT-5.2 的强大，然而，Google DeepMind 的工程师 @bcaine 直接用Gemini-3.0-pro 跑了一遍，碾压了。我按照他的方法复现了一下，果然没毛病。图片是我复现的结果。可以看到 Gemini-3.0-pro

>be David Budden >former senior engineer at DeepMind >quit to start startup PingYou >first rules of startup: “make what people want, talk to users” >$45,000 in public bets to solve Navier-Stokes >chump change for DeepMind salary >viral on X from AI mathtwit nerds >everyone

Google DeepMind researchers don't prompt like everyone else. I reverse-engineered their "role reversal" technique from leaked research papers. The difference is insane. 40% accuracy boost on logical reasoning. Here are the 5 insider methods they don't want you to know: